Urologic Oncology

SLIDES & TRANSCRIPTS
Saturday, December 6, 2003

Surrogate and Intermediate End Points for Prostate Cancer Specific Mortality

Anthony D'Amico, M.D., Ph.D.

Slide 1:

I'd like to thank Dr. Klein, Dr. Linehan and Dr. Thompson for the invitation to present. I always like to start with a brief story to set the stage, and the stage will be, how good does good enough have to be to make it over the mark that we use to define end points.

The story is this. There are a lot of octogenarians, and there is this woman, age 81, sitting around and talking with this group of men, all of a similar age group. She has got her two hands closed like this, and she says to the group of men, if you can guess what is in my left hand, then you can spend the night with me and have sex with me.

One man excitedly jumps up and says, I know, I know, it is an elephant. She looks over at him and smiles and says, close enough. I think our bar may be a little bit higher.

Surrogate and intermediate end points for cancer specific mortality.

Slide 2:

The background has been set by studies from Johns Hopkins, UCLA and the Mayo Clinic, in which they have shown, the first paper being the Pound paper in JAMA in '99, that the PSA doubling time following surgery -- and there are three studies, from Michigan, M.D. Anderson and Fox Chase -- following radiation is the primary determinant of time to the appearance of bone metastases on bone scan.

The interesting point is, it is independent of the hormonal therapy that is initiated at the time when the bone scan is positive, as in the Pound paper from the Hopkins group, or whether it is after PSA failure and some time before the bone scan is positive, such as the Mayo Clinic and the UCLA data.

So the clinical implication of that finding from those studies is that a short PSA doubling time following surgery or following radiation implies that there is metastatic disease, occult, just not detectable by our radiologic studies.

Slide 3:

The next two slides are a pilot study which was published in JCO at the end of 2002. Around that time, studies on cancer specific mortality were just beginning to emerge. This is a study where we looked at a small number of men, 394, treated with radiation and ask the question, does the doubling time after radiation predict for cancer specific death.

You can see here stratified cancer specific mortality by a doubling time more than a year, less than a year. The solid curve at the bottom is the group less than a year. Obviously they are doing more poorly than those with longer or more protracted rises.

Slide 4:

The interesting finding is the next slide. I draw your attention to this curve of the short doublers. Here is overall survival. This curve here has essentially remained unchanged. The curve of the people with the slowly rising PSA dropped considerably because they had time to die of other causes, since this is overall survival, while their PSA was slowly rising. But the people who were rising quickly, if they died, they died of prostate cancer.

Slide 5:

This raised the hypothesis that a short post treatment PSA doubling time following surgery or radiation may act as a surrogate for cancer specific death.

A word at this point is, why even pursue such a topic? The reason is for clinical trial design; end points need to be considered. A surrogate end point does two things to clinical trial design. It shortens the time to which the end point is achieved or can be evaluated, and it decreases markedly the number of patients that need to be randomized to answer a clinical question.

Slide 6:

The criteria that we used to assess that hypothesis was Prentice's criteria, which has two requirements, one, that the surrogate needs to be a prognostic factor, which means that depending on the value, the time to death following achievement of the surrogate needs to vary.

But the second one which is more difficult, is that the initial treatment received needs to be independent, if you will, of time to cancer specific death once the surrogate is achieved, which means that whether you are operated on or had radiation or any form of local therapy, for that matter, it didn't matter, once you achieved the surrogate, the time to death. In other words, whatever treatment you had subsequent or prior to the achievement of the surrogate had no impact on that end point.

Slide 7:

To this end, we went to Dr. Carroll and Dr. Moul from CapCure and CPDR database, and we combined these databases to come up with a little over 8,600 patients to answer this question in a surgically and radiation managed cohort.

I am skipping some slides so we can do this in the allotted time. These are the results from the Cox regression. In the doubling time group less than three months, when looked at as a categorical variable, doubling time was a prognostic factor with high significance. But when looking at treatment as a covariate in the Cox regression model, that is, whether you had surgery or radiation, if you achieved the surrogate, then the initial treatment and any subsequent treatment had no impact on time to cancer specific death.

The interesting finding to me in this study is that when you looked at a doubling time more than three months, if you looked at it as a continuous variable, it too was a prognostic factor, and treatment again made no difference once you knew the value of the doubling time when it was three months or more.

Said another way, given the value of the doubling time of three months or more, the time to death was defined by the value of the doubling time irrespective of initial treatment. For any value of doubling time of less than three months, it didn't matter if it was one, two or three months; people all died at the same time to cancer specific death, being a median of six years in this category. So here, the value matters, here it doesn't, but in either event, it can be used as a surrogate based on the criteria shown here.

Slide 8:

This is the results shown graphically. Here is doubling time less than three months in surgical patients, radiation patients, and you can see that these are not different. The median survival is six years following PSA failure, which shows you a very poor prognostic group, unlike the Pound paper, which had a median survival of 13 years overall following PSA failure for all comers. If you group all these patients together, you will find something similar.

Here is doubling time three months or more, surgery and radiation managed patients doing extremely well. What is not shown here, but I have other slides on, if you go to six months or less, nine months or less, 12 months or less, when you get up to 12 months or less, you are looking like this. No one is dying of prostate cancer in the first decade following PSA failure. The death rates are in the single digits. But the six-nine are interspersed in here. So one could choose a value based on the clinical trial you wanted to design and the patient population you wanted to study based on median survivals shown here.

Slide 9:

So the summary here and this was addressed in the editorial where this work appeared in JNCI in September, is as follows. It appears that PSA failure itself is not a surrogate. There are two large pieces of information that support that. One is the database I just presented, which is a compilation of CapCure and CPDR. But we applied parenthesis criteria to PSA failure in addition to doubling time.

We found that time to PSA failure was a significant predictor, was a prognostic factor for time to cancer death, but that initial treatment was also significant, with a hazard ratio of 2.3, showing that surgery was more favorable than radiation. This violates Prentice's criteria and says that you cannot use PSA failure as a surrogate for cancer specific death.

The RTOG did the same thing, Howard Sandler's group, looking at radiation with short versus long term hormones, a randomized trial published in JCO last month, found the same thing; time to PSA failure was a significant predictor, but that initial treatment also was a significant predictor, and therefore violated Prentice's criteria.

It is a commonsense conclusion, but now shown with a randomized and a large cross-sectional database, that PSA failure is not a surrogate for cancer specific death following surgery or radiation. But I believe, as you can see from the data I presented, doubling time is. I think the reason is because it reflects the underlying pace of the disease, volume or bulk of metastatic disease, or in other words, the biology.

Slide 10:

So in summary for this first part, we have shown that the doubling time following surgery or radiation from the Hopkins group, UCLA and Mayo Clinic, is the primary determinant of the time interval to a positive bone scan following PSA failure, independent of a time for hormonal therapy. The further data that we have shown is that it appears to be a surrogate for cancer specific mortality.

The clinical implication, and where do we take it, is that if you have a short post treatment PSA doubling time following surgery or external beam radiation, it implies that current methods of treatment are inadequate to cure that particular patient. In fact, they have currently lethal or incurable disease. Hopefully, that statement will change as new trials appear and new drugs and systematic therapies appear to address this population which has micro metastatic disease that is typically occult.

Slide 11:

Now, the last part. Going beyond the de novo patient who is getting surgery or radiation, what about the patient with a rising PSA following local therapy? Do we have anything that suggests that there may be an end point that is intermediate between the rising PSA and death that might be useful for ending clinical trials or studying new systemic therapies in clinical trials?

Here is a typical patient whose PSA is rising. Goes on hormonal therapy, PSA falls. We have always assumed that PSA falls in everyone following salvage hormonal therapy. While that is true, people go in at different rates, they have different doubling times entering hormonal therapy, and they exit at different rates.

We looked therefore at something we defined as PSA response, which is the ratio of the post or pretreatment PSA slope around hormonal therapy, and asked the question as to whether or not that in fact could be used as an intermediate end point to predict cancer specific death.

Slide 12:

It means that if you have a slow rise and a quick fall that is a large number over a small number, the absolute value of that number will be large, a good response. If you have a fast rise, a short doubling time and a slow fall, you would have a small number over a large number and the PSA response would be poor, or a low value of PSA response.

Slide 13:

For this, we went to a single institution cohort where I work of radiation managed patients. We had 199 men who had a rising PSA with bone scan negative, who underwent hormonal therapy, and we used as a validation cohort about 1200 men from the CapCure CPDR database from 44 institutions around the country

Slide 14:

to test the hypothesis that PSA response can predicted time to cancer specific mortality following salvage hormonal therapy. Cox regression analysis was used, with the ratio and the components of the ratio evaluated. We wanted to assess whether the ratio itself had any prognostic significance for predicting cancer specific death.

Slide 15:

The results show that the ratio is an important factor, both in the study and validation cohort. The pre and post treatment slope was not. Interestingly, the pretreatment PSA doubling time was.

Here is a patient with PSA doubling time of less than three months, three months or more, stratified by the response, from poor response to better response. Here is cancer specific death five years following hormonal therapy. You can see as you go from poor response to better response, there is a gradation in both subsets, obviously with more deaths occurring in the poor cohort, with shorter doubling times.

Slide 16:

Graphically, here is the study cohort stratified by a poor PSA response and a good PSA response, and that is statistically significant different,

Slide 17:

as is in the validation cohort.

The absolute numbers of deaths here are different between the study and validation cohort, because the study cohort are all radiation managed patients, therefore much poorer prognostic factors up front at presentation, whereas the validation cohort has surgical and radiation managed patients having a more favorable subset overall. Therefore, the absolute death rates are not as large here in the poor PSA response group, but still statistically significantly different than the good PSA response group.

Slide 18:

Finally, the interesting group, and where I will end my comments, is the doubling time less than three month group. There is a significant difference in the PSA response in these patients. These are the patients that I would submit we need to study in clinical trials.

Slide 19:

I would suggest the following clinical trial design, where you accrue patients in the rising PSA setting whose doubling time is short; they are randomized to hormonal therapy with or without novel therapy. I think we can justify hormonal therapy in this cohort because the median survival is short following PSA failure. Their time to positive bone scan is about a year and a half following PSA failure, so there is not a long time before they will be on hormonal therapy anyway for clinical symptomatic progression. I would submit that the end point you could look at would be PSA response.

Slide 20:

So the final summary slide is that this response parameter, which is the ratio of post to pretreatment slope in the rising PSA patient, has been validated as a prognostic factor following salvage hormonal therapy. It provides evidence in my mind to support the use of PSA as an intermediate end point in clinical trial designs in the rising PSA cohort, and the clinical implication which is all important, because that would enable drug discovery, like all the things we just talked about in the last session, to occur in a more timely fashion in the patients with a short doubling time, that we know today are currently incurable.

Thanks for your attention.